The general approach

I was interested in comparing aerial count detections to camera trap detections during the same period (late dry season of 2016). I was curious to see which species were more commonly detected in one vs. the other, and how the ratio of aerial to camera trap detections varied across habitats and species.

I cropped the 2016 aerial count data to the camera trap grid. For each species in the aerial count, I tallied up the number of individual animals counted in each of the 60 hexagonal grid cells. (I also counted the number of groups, but am not showing those analyses below, because Marc Stalmans informed me that multiple groups were often lumped together during the count for the sake of efficiency, particularly for common species. Also, the individual vs group aerial counts are highly correlated, so the patterns were similiar regardless of which was used as an aerial count measure.)

Here is a map showing the aerial count records within the grid:

I then calculated the Relative Activity Index, a simple measure of camera trap detections per night. Here, I considered records to be independent if they were >10 minutes from a record of the same species at that location. I did not record the number of individuals in each group, so in this case, we are looking at groups—an imperfect comparison with individuals from the aerial count, but the best that we can do so far. Meredith is currently working on determining the number of individuals in each record.

The aerial survey took place from October 18-31, 2016. I calculated RAI for three different time periods:

For each of these three time periods, I calculated RAI for each species:

Detections within each grid cell with aerial vs. camera surveys

First, I looked at whether or not each species was detected in each of the 60 grid cells in the aerial and/or camera trap records. Note that this list only includes species detected in the aerial survey. There were a number of species that were not counted in the aerial survey but which were present in the camera traps (including all of the carnivore species).

These figures indicate the number of grid cells in which the species was detected in both aerial and camera surveys (YAer_Y_Cam), in neither survey (NAer_N_Cam), ONLY camera trap (NAer_YCam), and ONLY aerial (YAer_NCam). The top plot is based on the 2-week camera survey, and the bottom on the 10-week survey. As would be expected, the detections of species increase over a longer survey period.

## Saving 7 x 5 in image

## Saving 7 x 5 in image

## Saving 7 x 5 in image

Some observations based on the 10-week period:

Species that are more often detected on aerial surveys but missed on camera traps include:

Some traits shared by many of these species include their medium/large body size, group formation, and preference for open habitat. These traits may make them more likely to be detected in the aerial survey.

Species that are more often detected on camera traps but missed on aerial surveys include:

These species tend to be more solitary and nocturnal, with smaller bodies, and found in more densely-wooded areas. As a result, they may be missed during the aerial survey.

Note that the observed elephant pattern is likely because they are highly mobile and are counted across multiple camera grid cells in the 10-week period.

Waterbuck and warthog are ubiquitous and nearly universally-detected (though as discussed below, many more warthogs are picked up in the camera traps than aerial survey).

Comparing relative number of aerial and camera detections within grid cells

Here is the correlation between the aerial count (number of individuals) and camera trap RAI, where each point represents one of the 60 grid cells. These plots are on a log scale for easier interpretation. This plot is shown for the 2-week window that occurred only during the aerial count, but the patterns look similar for the 6- and 10-week periods also.

## Saving 7 x 5 in image

## Saving 7 x 5 in image

## Saving 7 x 5 in image

## Saving 7 x 5 in image

## Saving 7 x 5 in image

## Saving 7 x 5 in image

You see that there is generally a positive relationship between detections on the aerial count and in the camera trap, though not a very clean relationship. It appears stronger for some species (waterbuck, warthog, impala) and weaker for others (nyala, bushbuck, reedbuck).

Comparing aerial and camera detections for different species

Then, I looked at overall correlation between aerial survey count and camera trap detections across the entire study area (all grid cells combined). I was interested in seeing how the ratio between the two counts varied among species. The size of the point corresponds to the size of the animal.

It appears that larger-bodied animals are relatively more detected in the aerial counts, as compared to the camera traps. Another way of looking at the same thing: here is the ratio of aerial count to RAI, as a function of body weight. A higher ratio means that the species was relatively more detected on the aerial counts.

## Saving 7 x 5 in image

## Saving 7 x 5 in image

## Saving 7 x 5 in image

So species like buffalo, sable, reedbuck, waterbuck, elephant, impala are picked up relatively more on the aerial surveys. Common duiker, red duiker, baboon, bushpig, bushbuck, warthog are picked up relatively more on the camera traps. This pattern of relative detection rates echoes what was seen above with the presence/absence comparison.

Detection and tree cover

I was curious to know if the tree cover in a given grid cell changed the relative patterns of detection based on the aerial count vs. camera traps, with the hypothesis that camera traps would pick up more animals in wooded areas where visibility from helicopter may be lower.

I only just began these analyses; ideally, I would want to calculate average tree cover within the entire grid cell, but this analysis was very computationally-intensive given the high-resolution rasters, so for now I just took the tree cover value at the camera location in the center of the grid cell.

Another important caveat: this only includes grid cells in which the species was detected in both aerial and camera surveys (since ratio would be 0 or infinity, which was messing up the regression). So interpret with caution.

And given all of those caveats, the initial graphs don’t show much. It seems that tree cover has little bearing on the relationship between aerial vs. camera detections, but again, this is preliminary.

## Saving 7 x 5 in image

## Saving 7 x 5 in image

## Saving 7 x 5 in image